Re: Minority objection to requiring unique GEDs or required feature to distinguish operations

On Fri, Aug 13, 2004 at 12:21:51PM -0400, Amelia A Lewis wrote:
> > Could you describe in detail what these customers of yours are trying to
> > do please?
> 
> Sorry, no.  One of them I can't describe, because I don't understand how
> it works.  The other two I've described approximately.  I'll try, without
> breaching confidentiality, to describe a little more.

Heh, of course I wasn't asking you to breach confidentiality, as much
fun as that might be. 8-)

> It's interesting that there are two of them, actually.  One of them uses a
> pretty easy-to-describe algorithm; there's a field (pretend it's called
> "command") in each message that identifies it.  Always the same xpath
> (this is rpc in wsdl 1.1, so better to say always the same xpath apart
> from the root element, taking the first child of the body as the root
> element).
> 
> It wouldn't be too hard to generate a URL that describes this.

Right.

> Now, the other uses multiple fields, and the second (and greater than)
> fields depend upon the content of the first field.  That is, one
> particular operation may be identified by the value "1" in field "root/x"
> plus the value "7" in field "/root/a/b/c", whereas another operation is
> selected by the value "2" in field "root/x" and the value "4" in field
> "root/l/m".  Actually, it's more complex than that, but that's the general
> idea.

Ok, but the key question is, does the algorithm ever use information
that isn't in the message itself (note, by "message" I mean all
information sent between agents, including, for example, TCP/IP and/or
underlying protocol data)?

If it doesn't - which appears to be the case as you've described it -
then you're good (see below).  If it does, then meaningful communication
cannot occur since there's an ambiguous contract in effect between the
agents.

> Question: should there be one URL that says "operation identifier
> contained in message"?  Or does every XPath need a different URL?  Or is
> there some middle ground, such as a URL per algorithm to determine the
> location of the operation indicator?

Bingo!  Yes, URI-per-algorithm is exactly what you want, I believe.

>  Note that, in no case are any of
> these "operation names" in the WSDL sense; it is possible to build a table
> that maps these indicators (singular or plural per message) to an
> operation.  Big table, in two cases (one I haven't described; I assume
> it's table-driven).

Right.  The URI just identifies an algorithm for finding some token
which can be unambiguously mapped to the operation (it needn't be an
"operation name" ala "getStockQuote", but could just be "1" which means
the same thing per the specifications describing the operational
semantics).

Mark.

Received on Friday, 13 August 2004 16:57:12 UTC